Ali BORJI, University of Wisconsin-Milwaukee, USA
Neil D. B. BRUCE, University of Manitoba, Canada
Ming-Ming CHENG, Nankai University, China
Jian LI, National University of Defense Technology, China
To be announced
Course Motivation and Description
Recently, visual saliency has received extensively growing attention across many disciplines including cognitive psychology, neurobiology, image processing, and computer vision. Based on our observed reaction times and estimated signal transmission times along biological pathways, human attention theories hypothesize that the human visual system processes only parts of an image in detail, with only limited processing of areas outside of the focus of attention. From an engineering perspective, such visual attention mechanisms have inspired a series of key research topics in the last few decades. One of the key forces behind these rapid developments is the vast amount of successful applications. These applications, marked by different requirements and points of emphasis have resulted in a rich kinship between fixation prediction, salient object detection, and objectness proposal generation.
It is noted that there has consistently been many papers about visual saliency appearing in ICIP over the past decade. While there are still many open issues and challenges (sometimes diverging arguments and debates) that need to be addressed in this area, the field of saliency computing continues to grow very rapidly. In this tutorial, we will introduce basic ideas, important models and applications of visual attention and saliency. Some key research issues will be discussed including top-down vs. bottom-up attention, and the relationship between fixation prediction, salient object detection, object proposal generation, etc. Recent advances in fixation prediction, salient object detection, and objectness proposals will be introduced in detail, with a significant emphasis on their respective potential applications. Finally, we will discuss the fairness of model evaluation criteria, model benchmarking, divergent opinions, open challenges, and potential future work.
This tutorial will consist of 5 talks (about 35-40 minutes for each talk). This begins with the fundamental knowledge and important classical models. Then, we discuss the divergence of, and correlation among different subareas (fixation prediction, salient object detection, and objectness proposals), followed by detailed introduction to each subarea. Finally, we discuss topics relating to model evaluation and benchmarking. The contents of the tutorial are as follows.
- - Fundamentals of visual attention and saliency and some important models. [Dr. Bruce]
- - Top-down vs. bottom-up attention, relationship between fixation predictions, salient object detection, object proposal generation, etc. [Dr. Borji]
- - Recent advances in fixation prediction, evaluation metrics and ground truth, and potential applications. [Dr. Jian]
- - Recent advances in salient object detection, and objectness proposals, and potential applications. [Dr. Cheng]
- - The fairness of model evaluation criteria (for both fixation prediction and salient regions detection) and model benchmarking. [Dr. Borji]
The attendee only needs to have basic knowledge of digital image processing in order to follow the course.
All materials will be distributed to the attendees electronically via webpage downloads. No physical materials will be distributed.
Ali BORJI received his B.S. and M.S. degrees in computer engineering from the Petroleum University of Technology, Tehran, Iran, 2001 and Shiraz University, Shiraz, Iran, 2004, respectively. He received his Ph.D. degree in computational neurosciences from the Institute for Studies in Fundamental Sciences (IPM) in Tehran, 2009. He then spent a year at University of Bonn as a postdoc. Before coming to the University of Wisconsin-Milwaukee in the fall of 2014, Dr. Borji was a postdoctoral scholar at iLab, University of Southern California, Los Angeles for four years.
Ming-Ming CHENG is an associate professor with College of Computer and Control Engineering, Nankai University. He received his PhD degree from Tsinghua University in 2012 under guidance of Prof. Shi-Min Hu, and working closely with Prof. Niloy Mitra. Then he worked as a research fellow for 2 years, working with Prof. Philip Torr in Oxford. Dr. Cheng’s research primarily centers on algorithmic issues in image understanding and processing, including image segmentation, editing, retrieval, etc. During the past 5 years, he has published a serials of influential papers in several sub-areas of visual saliency modeling, including salient object detection (e.g. his CVPR 2011 paper has received 790+ citations), objectness estimation (e.g. his CVPR 2014 oral paper has received 70+ citations and 3000+ source code downloads), and visual saliency based applications (e.g. his SIGGRAPH Asia 2009 paper ‘Sketch2Photo’ has received 250+ citations, and been reported by ‘The Telegraph’ from UK and ‘Spiegel’ from Germany).
Neil D. BRUCE is an Assistant Professor at the University of Manitoba in Canada. His research interests include a variety of topics including both computer vision and human vision, image processing, visual attention, machine learning, computational neuroscience, information theory, sparse coding, 3D modeling and reconstruction, natural image statistics, and statistical and graphical models. Prior to joining the University of Manitoba he completed two post-doctoral fellowships, one at the Centre for Vision Research at York University, and the other at INRIA Sophia Antipolis. Previously, he completed a Ph.D. in the department of Computer Science and Engineering in 2008 as a member of the Centre for Vision Research at York University, Toronto, Canada. In 2003, he completed a M. A. Sc. in System Design Engineering at the University of Waterloo, and received an Honors B.Sc. with a double major in Computer Science and Mathematics from the University of Guelph in 2001.
Jian LI is an assistant professor with National University of Defense Technology. He received the B.E. degree, the M.E. degree and the PhD Degree from National University of Defense Technology (NUDT), Changsha, P.R. China. From Jan 2010 to Jan 2011, he was a visiting Ph.D. student (Academic Trainee) at Center for Intelligent Machines (CIM) in McGill University under the supervision of Prof. Martin Levine.
Afternoon Sessions (13:30-17:00)
TPM-T1 (Invited) – Computational Photography +
Mohit GUPTA, Columbia University, USA
Jean-François LALONDE, Université Laval, Canada
To be announced
Course Motivation and Description
In the last decade, computational photography has emerged as a vibrant field of research. A computational camera uses a combination of unconventional optics and novel algorithms to produce images that cannot otherwise be captured with traditional cameras. The design of such cameras involves the following two main aspects:
- Optical coding – modifying the design of a traditional camera by introducing programmable optical elements and light sources to capture maximal amount of scene information in images;
- Algorithm design – developing algorithms that take information captured by conventional or modified cameras, and create a visual experience that goes beyond the capabilities of traditional systems.
Examples of computational cameras that are already making an impact in the consumer market include wide field-of-view cameras (Omnicam), light-field cameras (Lytro), high dynamic range cameras (mobile cameras), multispectral cameras, motion sensing cameras (Leap Motion) and depth cameras (Kinect).
This course serves as an introduction to the basic concepts in programmable optics and computational image processing needed for designing a wide variety of computational cameras, as well as an overview of the recent work in the field.
- A brief history of photography − Camera Obscura − Film, Digital and Computational photography;
- Coded photography − Novel camera designs and functionalities, including:
- - Optical coding approaches: Aperture, Image plane, and Illumination coding; Camera arrays,
- - Novel functionalities: Light field cameras − Extended DOF cameras, Hyperspectral cameras − Ultra high-resolution cameras (Gigapixel) − HDR cameras − Post-capture refocusing and Post-capture resolution trade-offs,
- - Depth cameras: Structured light − Time-of-flight,
- - Compressive sensing: Single pixel and High speed cameras;
- Augmented photography: algorithmic tools for novel visual experiences:
- - Multiple viewpoints: Image stitching, panoramas − Gigapixel imaging − Large-scale structure from motion,
- - Data-driven approaches: Texture transfer − Object transfer − Color/attribute/style transfer,
- - 2D image plane vs 3D scene: Scene geometry estimation − Light, geometry, and object editing,
- - Smarter tools: Content-aware inpainting − Edit propagation in image collections − Matte cutouts,
- - Smartphone photography: Cheap optics / powerful computing − Virtual tripod, Burst-mode HDR and denoising − Video stabilization,
- - Motion magnification and visual microphone;
- Future and impact of photography:
- - "Social/collaborative photography" or the Internet of Cameras,
- - Wearable and flexible cameras,
- - Seeing the invisible: seeing around corners, through walls, laser speckle photography,
- - Image forensics,
- - Next generation applications (personalized health monitoring, robotic surgery, self-driving cars, astronomy).
Basic knowledge of linear algebra and probability.
Course PowerPoint / keynote slides.
Jean-François LALONDE is an assistant professor in Electrical and Computer Engineering at Laval University, Quebec City. Previously, he was a Post-Doctoral Associate at Disney Research, Pittsburgh. He received a B.Eng. degree in Computer Engineering with honors from Laval University, Canada, in 2004. He earned his M.S at the Robotics Institute at Carnegie Mellon University in 2006 and received his Ph.D., also from Carnegie Mellon, in 2011. His Ph.D. thesis won the 2010-11 CMU School of Computer Science Distinguished Dissertation Award, and was partly supported by a Microsoft Research Graduate Fellowship. After graduation, he became a Computer Vision Scientist at Tandent, where he helped develop LightBrush™, the first commercial intrinsic imaging application, and introduced the technology of intrinsic videos at SIGGRAPH 2012. His work focuses on lighting-aware image understanding and synthesis by leveraging large amounts of data. More details about his research can be found here.
Mohit GUPTA will start as an assistant professor in the CS department at the University of Wisconsin-Madison in January ’16. He is currently a research scientist in the CAVE lab at Columbia University. He received a B.Tech. in computer science from Indian Institute of Technology Delhi in 2003, an M.S. from Stony Brook University in 2005 and a Ph.D. from the Robotics Institute, Carnegie Mellon University in 2011. His research interests are in computer vision and computational imaging. His focus is on designing computational cameras that enable computer vision systems to perform robustly in demanding real-world scenarios, as well as capture novel kinds of information about the physical world. Details can be found here.
TPM-T2 – Example-based Super Resolution +
Jordi SALVADOR, Technicolor – Deutsche Thomson, Germany
Mehmet TURKAN, Technicolor, France & Izmir University of Economics, Turkey
To be announced
Course Motivation and Description
Super Resolution has been one of the most popular research disciplines in image processing during the last years. From the research perspective, the reasons for this success include the interesting solutions to combinations of different image processing problems (registration, deblurring, denoising…) or the increasing understanding of the subspace of natural images and its proper application in recent statistical models. Besides, the introduction of new imaging standards with progressively higher resolutions favors the interest on new upscaling algorithms also in the industry. When properly designed, super-resolution methods are capable of adapting legacy contents to the resolution offered by the latest display technologies, either during postproduction or directly at the end user’s devices, thus offering optimal visual experiences.
During the last years, research on example-based super resolution has received the main focus of attention essentially due to two reasons: In first place, in contrast with classic multi-frame super resolution, the use of more advanced image priors alleviates the requirement of having different captures of the same scene with subpixel shifts. Furthermore, numerical stability problems that might arise when reconstructing a super-resolved image under the commonly over-simplified parametric models in multi-frame super resolution are also avoided by using more meaningful non-parametric image priors.
This tutorial is designed to present an evolutionary timeline of the many existing and continuously improving state-of-the-art approaches that benefit from the favorable features of example-based super resolution, with insights on the theoretical background, implementation issues (including parallelization) and discussion on the practical applicability.
The tutorial provides a thorough introduction and overview of example-based super-resolution, covering the most successful algorithmic approaches, the theory behind them, implementation insights, and some hints about current challenges and expected outcomes for the near future. The list of covered topics is as follows.
- Introduction to super resolution
This section introduces early (non-example-based) super-resolution pipelines and the rationale of the example-based concept covered by the rest of the tutorial.
- - A historic view of super resolution
- - Multi-frame super resolution
- - Example-based super resolution
- Self-similarity-based super resolution
This part of the tutorial describes super-resolution models where examples are learned from one or more scales of the input data. This strategy can be efficiently implemented when hardware solutions for block search are available, and has the nice property of being implicitly adaptive to the input contents.
- - High-frequency transfer
- - Locally linear embedding
- - Robust self-similarity
- Super resolution by external learning
This section will cover super-resolution strategies where larger amounts of data can be exploited to build suitable regression models during an offline training stage. These models can then be efficiently applied during the online inference stage. Under proper configurations, the generalizability of these machine-learning approaches can be virtually as high as that of self-similarity-based approaches and the reconstruction quality is often superior.
- - Dictionaries
- - Anchored neighbors and variations
- - Hybrid models: self-similarity and regression
- - Regression trees
- - Deep learning
The attendees should be familiar with basic concepts in image processing, probability and statistics (undergraduate courses suffice), but the tutorial is self-contained for the most part.
All registered attendees shall receive printouts of the supporting slides.
Jordi SALVADOR is project leader at Technicolor R&I; in Germany, where he started working in 2011, and member of Technicolor’s Fellowship Network since 2014. His main research focus is on machine learning for example-based super resolution and image restoration. Formerly, he received a M.Sc. in Telecommunications (equivalent to Electrical) Engineering in 2006 and a M.Sc. in the European MERIT program in 2008, both from the Universitat Politècnica de Catalunya (UPC) in Barcelona. He obtained the Ph.D. degree in 2011, also from UPC, where he contributed to projects of the Spanish Science and Technology System (VISION, PROVEC) and to a European FP6 project (CHIL) as research assistant on multi-camera 3D reconstruction. He has also served as reviewer in conferences and journals like EUSIPCO and IEEE Transactions on Image Processing. His research interests include 3D reconstruction, real-time and parallel algorithms, new computer-human interfaces, image and video restoration, super resolution, inverse problems and machine learning.
Mehmet TÜRKAN is a researcher at Technicolor R&I; in Cesson-Sévigné, France, since 2011. He will be joining the Engineering and Computer Science Faculty of Izmir University of Economics, Izmir, Turkey, in Sept 2015. He obtained his PhD degree in computer science from INRIA-Bretagne Atlantique- and University of Rennes 1, Rennes, France. He received his MSc and BSc (Hhons) degrees both in electrical and electronics engineering from Bilkent University, Ankara, and Eskisehir Osmangazi University, Eskisehir, Turkey, respectively. He was involved with the European Commission (EC) 6th Framework Program (FP6) Multimedia Understanding through Semantics, Computation and Learning Network of Excellence (MUSCLE-NoE), EC FP6 Integrated Three-Dimensional Television–Capture, Transmission, and Display Network of Excellence (3-DTV-NoE), and European UltraHD-4U research projects. His general research interests are in the area of signal processing with an emphasis on image and video processing and compression, pattern recognition and classification, and computer vision. Dr. Türkan was the recipient of the Best Student Paper Award in the 2010 IEEE International Conference on Image Processing (ICIP) and was a nominee for the Best Student Paper Award in the 2011 IEEE ICIP.
TPM-T3 – Perceptual Metrics for Image and Video Quality in a Broader Context: From Perceptual Transparency to Structural Equivalence +
Thrasyvoulos N. PAPPAS, Northwestern University, Evanston, Illinois, USA
Sheila S. HEMAMI, Northeastern University, Boston, Massachusetts, USA
To be announced
Course Motivation and Description
We will examine objective criteria for the evaluation of image quality that are based on models of visual perception. Our primary emphasis will be on image fidelity, i.e., how close an image is to a given original or reference image, but we will broaden the scope of image fidelity to include structural equivalence. We will also discuss no-reference and limited-reference metrics. We will examine a variety of applications with special emphasis on image and video compression. We will examine near-threshold perceptual metrics, which explicitly account for human visual system (HVS) sensitivity to noise by estimating thresholds above which the distortion is just-noticeable, and supra-threshold metrics, which attempt to quantify visible distortions encountered in high compression applications or when there are losses due to channel conditions. We will also consider metrics for structural equivalence, whereby the original and the distorted image have visible differences but both look natural and are of equally high visual quality. We will also take a close look at procedures for evaluating the performance of quality metrics, including database design, models for generating realistic distortions for various applications, and subjective procedures for metric development and testing. Throughout the course, we will discuss both the state of the art and directions for future research.
This course will enable you to:
- - Gain a basic understanding of the properties of the human visual system and how current applications (image and video compression, restoration, retrieval, etc.) that attempt to exploit these properties.
- - Gain an operational understanding of existing perceptually-based and structural similarity metrics, the types of images/artifacts on which they work, and their failure modes.
- - Understand current distortion models for different applications, and how they can be used to modify or develop new metrics for specific contexts.
- - Understand the differences between sub-threshold and supra-threshold artifacts, the HVS responses to these two paradigms, and the differences in measuring that response.
- - Understand criteria by which to select and interpret a particular metric for a particular application.
- - Understand the capabilities and limitations of full-reference, limited-reference, and no-reference metrics, and why each might be used in a particular application.
- - Applications: Image and video compression, restoration, retrieval, graphics, etc.
- - Human visual system review
- - Near-threshold perceptual quality metrics
- - Supra-threshold perceptual quality metrics
- - Structural similarity metrics
- - Perceptual metrics for texture analysis and compression – structural texture similarity metrics
- - No-reference and limited-reference metrics
- - Models for generating realistic distortions for different applications
- - Design of databases and subjective procedures for metric development and testing
- - Metric performance comparisons, selection, and general use and abuse
- - Embedded metric performance, e.g., for rate-distortion optimized compression or restoration
- - Metrics for specific distortions, e.g., blocking and blurring
- - Metrics for specific attributes, e.g., contrast, roughness, and glossiness
- - Multimodal applications
- - Basic understanding of image compression algorithms
- - Background in digital signal processing and basic statistics: frequency-based representations, filtering, distributions.
- - Level: Intermediate
PDF of PowerPoint presentation
Thrasyvoulos N. PAPPAS received the S.B., S.M., and Ph.D. degrees in electrical engineering and computer science from MIT in 1979, 1982, and 1987, respectively. From 1987 until 1999, he was a Member of the Technical Staff at Bell Laboratories, Murray Hill, NJ. He is currently a professor in the Department of Electrical and Computer Engineering at Northwestern University, which he joined in 1999. His research interests are in image and video quality and compression, image and video analysis, content-based retrieval, perceptual models for multimedia processing, model-based halftoning, and tactile and multimodal interfaces. Prof. Pappas will be serving as Vice-President Publications, IEEE Signal Processing Society (2015-107). He has served as editor-in-chief of the IEEE Transactions on Image Processing (2010-12), elected member of the Board of Governors of the Signal Processing Society of IEEE (2004-06), chair of the IEEE Image and Multidimensional Signal Processing (now IVMSP) Technical Committee, technical program co-chair of ICIP-01 and ICIP-09, and co-chair of the 2011 IEEE IVMSP Workshop on Perception and Visual Analysis. He has also served as co-chair of the 2005 SPIE/IS&T; Electronic Imaging Symposium, and since 1997 he has been co-chair of the SPIE/IS&T; Conference on Human Vision and Electronic Imaging. Dr. Pappas is a Fellow of IEEE and SPIE.
Sheila S. HEMAMI received the B.S.E.E. degree from the University of Michigan in 1990, and the M.S.E.E. and Ph.D. degrees from Stanford University in 1992 and 1994, respectively. She was with Hewlett-Packard Laboratories in Palo Alto, California in 1994 and was with the School of Electrical Engineering at Cornell University from 1995-2013. She is currently Professor and Chair of the Department of Electrical & Computer Engineering at Northeastern University in Boston, MA. Dr. Hemami's research interests broadly concern communication of visual information from the perspectives of both signal processing and psychophysics. She was elected a Fellow of the IEEE in 2009 for her for contributions to robust and perceptual image and video communications. Dr. Hemami has held various visiting positions, most recently at the University of Nantes, France and at École Polytechnique Fédérale de Lausanne, Switzerland. She has received numerous university and national teaching awards, including Eta Kappa Nu's C. Holmes MacDonald Award. She will be serving as Vice-President Publications Products and Services, IEEE (2015). She was a Distinguished Lecturer for the IEEE Signal Processing Society in 2010-11, was editor-in-chief for the IEEE Transactions on Multimedia from 2008-10. She has held various technical leadership positions in the IEEE.
TPM-T4 – Spectral Methods in 3D Data Analysis +
Michael BRONSTEIN, University of Lugano, Switzerland & Perceptual Computing, Intel
To be announced
Course Motivation and Description
Over the last decade, the intersections between 3D shape analysis and image processing have become a topic of increasing interest in the computer graphics community. Nevertheless, when attempting to apply current image analysis methods to 3D shapes (feature-based description, registration, recognition, indexing, etc.) one has to face fundamental differences between images and geometric objects. Shape analysis poses new challenges that are non-existent in image analysis.
The purpose of this course is to overview the foundations of shape analysis and to formulate state-of-the-art theoretical and computational methods for shape description based on their intrinsic geometric properties. The emerging field of spectral and diffusion geometry provides a generic framework for many methods in the analysis of geometric shapes and objects. The course will present in a new light the problems of shape analysis based on diffusion geometric constructions such as manifold embeddings using the Laplace-Beltrami and heat operator, 3D feature detectors and descriptors, diffusion and commute-time metrics, functional correspondence, and spectral symmetry.
The course is divided in four sections, covering the topics listed below.
- Theoretical foundations
Diffusion operators, their spectral properties, Fourier analysis on manifolds, similarities to the classical case − Heat diffusion equation on a Riemannian manifold − The Laplace-Beltrami operator − Diagonalization of Laplacians, relation to joint approximate diagonalization problems − The fundamental solution based on the heat kernel − The discrete heat operator and its basic algebraic properties − Scale-space and heat diffusion − The diffusion and the commute-time distances.
- Shape representation
Manifold embedding using the heat operator − Relationship with Laplacian embedding and diffusion embeddings − Geometric and photometric diffusion − Local and global diffusion geometry − Feature detection and feature description − Heat and wave kernel signatures − Optimal spectral descriptors. Convolutional neural networks on manifolds − Volumetric vs surface diffusion.
Minimum-distortion similarity and correspondences − Functional correspondence, relation to sparse coding and matrix completion problems − Intrinsic symmetry detection − Shape retrieval, bag-of-feature methods − Benchmarks.
- Implementation and application examples
Live demos in MATLAB to exemplify the main concepts of the tutorial.
Basic knowledge of signal/image processing, Fourier analysis
Course slides will be available online.
Michael BRONSTEIN is a professor in the Faculty of Informatics at the University of Lugano (USI), Switzerland and a Research Scientist at the Perceptual Computing group, Intel, Israel. Michael got his B.Sc. in Electrical Engineering (2002) and Ph.D. in Computer Science (2007), both from the Technion, Israel. His main research interests are theoretical and computational methods in spectral and metric geometry and their application to problems in computer vision, pattern recognition, computer graphics, image processing, and machine learning. His research appeared in international media and was recognized by numerous awards. In 2012, Michael received the highly competitive European Research Council (ERC) grant. In 2014, he was invited as a Young Scientist to the World Economic Forum New Champions meeting, an honor bestowed on forty world's leading scientists under the age of 40. Besides academic work, Michael is actively involved in the industry. He was the co-founder of the Silicon Valley start-up company Novafora, where he served as VP of technology (2006-2009), responsible for the development of algorithms for large-scale video analysis. He was one of the principal inventors and technologists at Invision, an Israeli startup developing 3D sensing technology acquired by Intel in 2012 and released under the RealSense brand.
TPM-T5 – Sparse stochastic processes: A unifying statistical framework for modern image processing +
Michael UNSER, EPFL, Switzerland
To be announced
Course Motivation and Description
Sparsity and compressed sensing are very popular topics in image processing. More and more, researchers are relying on the related l1-type minimization schemes to solve a variety of ill-posed problems in imaging. The paradigm is well established with a solid mathematical foundation, although the arguments that have been put forth in the past are mostly deterministic. In this tutorial, we shall introduce the participants to the statistical side of this story. As an analogy, think of the foundational role of Gaussian stationary processes: these justify the use of the Fourier transform or DCT and lend themselves to the formulation of MMSE/MAP estimators based on the minimization of quadratic functionals.
The relevant objects here are sparse stochastic processes (SSP), which are continuous-domain processes that admit a parsimonious representation in a matched wavelet-like basis. Thus, they exhibit the kind of sparse behavior that has been exploited by researchers in recent years for designing second-generation algorithms for image compression (JPEG 2000), compressed sensing, and the solution of ill-posed inverse problems (l1 vs. l2 minimization).
The construction of SSPs is based on an innovation model that is an extension of the classical filtered-white-noise representation of a Gaussian stationary process. In a nutshell, the idea is to replace 1) the traditional white Gaussian noise by a more general continuous-domain entity (Lévy innovation) and 2) the shaping filter by a more general linear operator. We shall present the functional tools for the complete characterization of these generalized processes and the determination of their transform-domain statistics. We shall also describe self-similar models (non-Gaussian variants of fBm) that are well suited for image processing.
We shall then apply those models to the derivation of statistical algorithms for solving ill-posed problems in imaging. This allows for a reinterpretation of popular sparsity-promoting processing schemes—such as total-variation denoising, LASSO, and wavelet shrinkage—as MAP estimators for specific types of SSPs. It also suggests novel alternative Bayesian recovery procedures that minimize the estimation error (MMSE solution). The concepts will be illustrated with concrete examples of sparsity-based image processing including denoising, deconvolution, tomography, and MRI reconstruction from non-Cartesian k-space samples.
- - Classical reconstruction algorithms and the Gaussian hypothesis
- - Variational formulations: from l2- to l1-norm minimization
- - Compressed sensing
Part I: Statistical modeling
An introduction to sparse stochastic processes
- - Generalized innovation model
- - Statistical characterization of signals
Part II: Recovery of sparse signals
Reconstruction of biomedical images
- - Discretization of inverse problems
- - Generic MAP estimator (iterative reconstruction algorithm)
- - Applications: deconvolution microscopy, MRI, x-ray tomography
From MAP to MMSE estimation
- - MMSE estimation of Markov processes
- - Iterative wavelet-domain MMSE denoising
Basic knowledge of statistical signal processing (MAP estimation), optimization techniques (iterative algorithms), and functional analysis (Fourier transform, generalized functions, differential equations)
Copies of the slides
Complete lecture notes for the tutorial (and beyond) are available on the web at http://www.sparseprocesses.org
Michael UNSER is Professor and Director of EPFL's Biomedical Imaging Group, Lausanne, Switzerland. His main research area is biomedical image processing. He has a strong interest in sampling theories, multiresolution algorithms, wavelets, the use of splines for image processing, and, more recently, stochastic processes. He has published about 250 journal papers on those topics. He is the leading author of “An introduction to sparse stochastic processes”, Cambridge University Press, 2014.
From 1985 to 1997, he was with the Biomedical Engineering and Instrumentation Program, National Institutes of Health, Bethesda USA, conducting research on bioimaging and heading the Image Processing Group.
Dr. Unser is a fellow of the IEEE (1999), an EURASIP fellow (2009), and a member of the Swiss Academy of Engineering Sciences. He is the recipient of several international prizes including three IEEE-SPS Best Paper Awards and two Technical Achievement Awards from the IEEE (2008 SPS and EMBS 2010).